How much memory Minecraft modpacks actually use

How this was measured

Measured on 3 August 2026 on the same hardware that runs customer servers. Every pack went through the normal create-server path, with no hand-tuning. Each was booted twice, and readings were taken at idle, on a freshly generated world, with nobody connected. Peak is the highest memory the container reported during the run.

Nobody was playing. A populated world with chunks loading will sit higher than anything in this table.

Every reading

Sorted by headroom, tightest first. Headroom is how much of the container limit was still free at peak, rounded.

Loader, Minecraft version and mod count are hidden on narrow screens. Open this on a wider one to see them.

Why some rows show almost no memory used

A handful of packs report a peak of a few megabytes against a multi-gigabyte limit. Those servers stopped before they loaded their mods, so the reading describes a failed start and not the pack. They are kept in the table because removing them would quietly improve the averages.

The loader build installed is often not the one the pack asked for

For 71 packs we could compare the loader build written in the pack's own manifest against the build that ended up on disk. They disagreed for 49 of them.

Usually the installed build is newer and nothing bad happens. Sometimes it is older than the pack needs, and the server dies on a mixin error that reads like a broken mod. This is why we now pin the build from the manifest instead of letting the installer pick.

Showing the 49 packs where the two disagreed.

So how much RAM should I actually give my modpack?

Start from the pack's own recommendation, then watch whether it crashes. Our guide to server memory walks through the bands by pack size and player count, and explains when the problem is memory and when it is tick time pretending to be memory.

Is this still accurate?

The readings are from 3 August 2026. Modpacks update constantly, so treat a specific number as a snapshot of that version on that day. The pattern it shows holds up better than any individual row.

Why not just publish the pass or fail result for each pack?

Because it would be out of date. These boot results came from a sweep that led directly to a batch of fixes on our side, after which 74 of the 100 packs worked rather than 46. Publishing the old verdicts would describe a platform that no longer exists.

Can I use these numbers?

Yes. The table is published under CC BY 4.0. Link back to this page and please keep the measurement date attached, because a number without its date is misleading.

• How much RAM does a Minecraft server need?
• Why modpack servers crash on startup
• Modpack hosting